# NL2Shell Ollama Modelfile

#

# NOTE: Qwen3.5's hybrid DeltaNet architecture is not yet supported by Ollama's

# GGUF loader. Once Ollama adds qwen35 architecture support, this Modelfile will

# work with the GGUF files on HuggingFace:

#

# ollama create nl2shell -f Modelfile

#

# Until then, Vox CLI defaults to qwen2.5-coder:0.5b as a fallback.

# The fine-tuned model can still be used via Python transformers:

#

# from transformers import AutoModelForCausalLM, AutoTokenizer

# model = AutoModelForCausalLM.from_pretrained("AryaYT/nl2shell-0.8b")

#

# For v3 training, consider Qwen2.5-Coder-1.5B as an Ollama-compatible base.

FROM gguf/Qwen3.5-0.8B.Q4_K_M.gguf

SYSTEM "You are an expert shell programmer. Given a natural language request, output ONLY the corresponding shell command. No explanations, no markdown, no code fences, no comments. Just the raw command."

PARAMETER temperature 0.1

PARAMETER num_predict 256

PARAMETER stop "<|im_end|>"

TEMPLATE """<|im_start|>system

{{ .System }}<|im_end|>

<|im_start|>user

{{ .Prompt }}<|im_end|>

<|im_start|>assistant

"""